Feasibility of Res-UNet-based phase unwrapping in compression optical coherence elastography

Compression optical coherence elastography (OCE) images the biomechanical properties of tissue by analyzing phase differences, where phase unwrapping (PU) is a critical step to build the unwrapped phase differences. Traditional weighted phase unwrapping (WPU) algorithm provides high accuracy, but suffers from slow processing speed when handling large OCE datasets. We present a Res-UNet-based framework for phase unwrapping in compression OCE, coupled with a synthetic data-generation pipeline that embeds realistic OCE background noise into wrapped phase difference maps, thereby addressing the scarcity of reliable absolute phase data free from noise and wrapping artifacts for model training. We evaluated the proposed approach using phase difference maps from tissue-mimicking phantom and ex vivo human breast tissue OCE datasets. Comparative experiments with WPU algorithm demonstrated that the proposed method achieved comparable performance in high-signal OCT regions, improved processing speed by more than twofold. These findings demonstrate the feasibility and potential of the proposed Res-UNet framework for rapid phase unwrapping in compression OCE.